Aankirz

Zoya — AI skill for Claude Code

AI community

Voice-first AI agent that operates a Mac for blind and low-vision people.

How to install Zoya

This entry records only its repository, not the path inside it, so there is no exact command to give. Open Aankirz/zoya and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Zoya does

Voice-first AI agent that operates a Mac for blind and low-vision people. Strands Agents + Amazon Bedrock.

Alternatives in AI

  • Guizang Ppt Skill — AI-agent Skill for generating polished HTML slide decks: editorial magazine and Swiss layouts, image prompts 24.9k ★
  • Gptaku Plugins — A Claude Code plugin marketplace for people who want to become AI Native 1.1k ★
  • Zhigeng — 知更 — 本地 AI 的上下文与记忆层 389 ★

README

Zoya

**A voice-first AI agent that operates a Mac for blind and low-vision people.**

Say **"Hey Zoya"** and describe a goal — *"order my usual groceries"*, *"make a presentation on renewable energy"*, *"what's on my screen?"* — and Zoya does the whole task: opens apps, drives the browser, clicks and types, remembers preferences, runs helper agents in the background, and narrates everything with calm speech and a vocabulary of soft sounds. Before anything irreversible (paying, sending, deleting) it **always** asks for a spoken "confirm".

*VoiceOver tells you what's on the screen. Zoya does what you meant.*

Built for the **Vision OS** hackathon with the **Strands Agents SDK** and **AWS**.


Status

🚧 **Hackathon build in progress.**

Architecture at a glance

![Zoya architecture: voice on the Mac, Strands Agents orchestrator, Amazon Bedrock Nova with OpenAI fallback, AWS services](assets/architecture.png)

Diagram source: [`assets/architecture.mmd`](assets/architecture.mmd) (Mermaid).

**How one request flows**

  1. Hear. The wake word, end-of-turn detection and speech-to-text all run on the Mac. Your voice never leaves it.
  2. Route. Simple commands take a rule and finish in milliseconds, with no AI call.
  3. Think. A real task gets its own Strands Agents orchestrator. It plans with Amazon Nova on Bedrock and falls back to OpenAI, behind one provider adapter (zoya/models.py, MODEL_PROVIDER).
  4. Act. Skills drive the browser and the Mac. Anything that pays, sends, posts or deletes passes the safety gate first, and only your spoken "confirm" unlocks it.
  5. Speak. Zoya narrates each step with the Amazon Polly voice and soft earcons, and the overlay shows the same steps on screen.
Layer What powers it
Agent framework Strands Agents SDK: agents, tools, hooks, streaming, OpenTelemetry
Models Amazon Bedrock (Amazon Nova), with OpenAI as fallback
Voice Whisper on-de